How can I retrieve the original stored text using semantic search in n8n?

Below is an overview of how you can set up the process in n8n so that semantic queries match against stored content, yet the returned text is exactly as originally stored:

1. Storing Your Original Text

When you insert your texts into the vector database, make sure that the original content is kept intact. In n8n you can do this by using a node (like the Pinecone Vector Store node, if you’re using Pinecone) in Insert Documents mode to store your documents along with their computed embeddings. This ensures that while the text is transformed into high-dimensional vectors for search, the raw text remains unmodified.

2. Querying with Semantic Search

To perform semantic or similarity search based on context rather than exact wording, you can use one of the following approaches:

• Vector Store Retriever Node:
  This node is designed to fetch documents from the vector database by matching the embedded query with stored vectors. This will return the document that is most similar semantically while keeping the text exactly as stored.

• Vector Store Node in Get Many/Retrieve Documents Mode:
  You can also use the Vector Store node configured in Get Many to return several similar documents or in Retrieve Documents mode if you’re chaining it with additional nodes (like a chat workflow) for further processing.

3. Generating the Query Embeddings

Before querying the vector store, you’ll typically want to convert your incoming query into an embedding using a node like Embeddings OpenAI. This will translate your bot’s query—regardless of phrasing—into a high-dimensional vector which you then use as the search query.

Example Workflow

1. Store Your Data:
   • Use an insertion node (ex: Pinecone Vector Store node set to Insert Documents) to store your text and its corresponding vector representation.
2. Process the Bot Query:
   • Use the Embeddings OpenAI node to convert the bot’s query into a vector.
3. Retrieve the Document:
   • Feed this vector into the Vector Store Retriever node (or the Vector Store node in Retrieve Documents mode) to perform similarity search.
   • The node will return the document(s) whose stored embeddings most closely match the query vector. Since these nodes retrieve the text as stored, the output text remains unchanged.

By combining these components, your semantic search will match on context and meaning while always returning the original stored text—even if the bot’s query was phrased differently.

This approach allows your workflow in n8n to handle AI-driven semantic searches accurately, ensuring that the text with variables is preserved in its original form during retrieval.